What’s left for us to do?

I hit enter, and the Battlecode bot appeared. Every line. Every logic gate. Every coordinate. I didn’t touch a single line of code.
It’s a far cry from when I started my SC2 bot only a few years ago. Combing through the API docs for hours, wrestling with Python, and the white-knuckle frustration of that bug that took three days to kill. Back then, the triumph felt earned because the struggle was manual. Now I just prompt it into existence, and think about it… this is the worst it’ll ever be.
I’d bet you’ve had that same hollow feeling. Spending an hour prompting a feature that used to take a week, and you realize your mastery has just become a way to check the AI’s homework.
It leaves you asking: if the machine handles the “how” perfectly, what’s actually left for us to do?

The panic comes from thinking that coding is the value. If you define yourself by your ability to write a clean loop or optimize a pathing algorithm, then yes, you’re in trouble. The AI already won that game.
The value has shifted from the “how” to the “what.”
I’ve stopped trying to compete with the LLM on syntax and started focusing my study on the things the AI can’t actually reason through. If you want to stay relevant, you’re going to have to stop being a bricklayer of code and start being an architect of systems.
I’ve narrowed my focus to four specific areas
1. Systems Thinking: The AI can write a beautiful function, but it often misses the big picture, specifically the way the economy, the unit composition, and the map geometry all lean on each other. That’s why I’ve shifted my focus to the whole system rather than the individual parts. It’s about managing the tension between those variables to ensure the bot doesn’t just execute well, but actually makes sense in the context of the game.
2. Game Theory: Coding a bot to attack is a solved problem, but figuring out when to attack based on how an opponent’s likely to react? That’s where LLMs fall super short. Talking to Zerospace’s creator made me realize that this is where “you” actually live in the code. It’s the creative imprint you leave on the bot. While the AI can calculate the odds, it can’t define the strategy. It can’t study the incentives, the risks, and the payoffs the way you can.
3. Utility Theory and Reward Shaping: This is a different set of skills entirely, where the goal is to turn an abstract concept into a concrete objective. You’ve got to tell the bot what “winning” actually looks like and how to choose between competing priorities, because LLMs tend to have an absurd, sometimes downright lazy view of success when it’s ill-defined. That’s why you can’t just let the AI shape the reward on its own; without a steady hand, you risk poisoning your whole project. Your real job here is to ensure the bot’s intelligence actually aligns with the goal.
4. Domain Knowledge: I use AIs to process my losing games all the time, and while the machine has every single coordinate and timestamp, it still hits a wall. It comes to these over confidently dumb conclusions because it skips the “unseen variables” the things that aren’t in the logs, like the vibe of the match, the psychological pressure of a specific play or straight up misinterprets what it processed.
That’s where your ultimate leverage lives. It’s the ability to look at an AI’s prescription and say confidently: “No, that’s not it. You’re wrong here.”
Because the AI is a powerful engine that’s completely blind, it can lay the bricks perfectly, but it doesn’t know if it’s building a wall or a bridge. You’re the only one who can provide the eyes.
⌨️ Next Commit
Saving a list of resources is just another way to feel productive without actually building anything. I’ve found that the only way to learn this stuff is “just in time,” where you only open the book when a real problem forces you to.
So here’s the move for this week. Pull up your bot’s ugliest recent loss and spend 15 minutes on one question: which of the four areas does the failure actually live in?
The answer is the only door you open.
- Economy, army, and map fighting each other? Thinking in Systems is the start.
- Bot attacking at all the wrong moments? Thinking Strategically is how you reason about the opponent, not just the board.
- Bot “winning” stupid? This GDC talk on utility theory rewired how I score every action.
- The loss makes no sense in the logs? Find a YouTuber who breaks down the game in an analytical way. Search for your specific matchup and watch how they explain the human reasoning behind certain actions. If you’re on SC2, I recommend PiG.
(Those book links are affiliate ones. Costs you nothing extra, throws a few cents back to the newsletter if you grab one.)
May the Bugs Be Ever In your Favour🪲
Email Preference:
Unsubscribe | Update your profile | 600 1st Ave, Ste 330 PMB 92768, Seattle, WA 98104-2246
